Image-based Rail Surface Defect Detection And Classification Algorithm Research

Rail surface defect detection is a research hotspot in the large-scale development of modern railways.Accurate and timely detection of defects can not only reduce the operating cost of the railway,but also ensure its driving safety.Commonly used rail surface defect detection methods mainly include ultrasonic detection,eddy current detection and visual observation.But they cannot meet the timeliness and robustness of rail surface defect detection.Image processing technology is one of the main technologies in the field of computer vision.It has merits of high efficiency,high precision and non-contact.In recent years,image processing technology has been introduced into the research of rail surface defect detection,which effectively compensates for its shortcomings.However,in the actual detection process,the traditional image processing algorithm can not accurately detect the surface defect image of the rail.The detection target is easily affected by the surrounding environment and the result of missed detection or false detection.Besides,the running speed of the whole detection system needs to be improved.Based on the related researches,the thesis conducts in-depth research on image-based rail surface defect detection and classification.Corresponding improved algorithms are proposed for existing problems.The main research results of this thesis are as follows:(1)A modified Canny based rail surface defect detection algorithm is proposed.The algorithm proposes an adaptive selection filtering window and an adaptive weighting coefficient algorithm to replace the Gaussian filtering in the traditional Canny algorithm according to the noise density in the rail surface image;the double threshold determined by minimizing the gradient based on the gradient magnitude histogram combined with the intra-class variance The evaluation function is combined with the OTSU algorithm to determine the upper and lower thresholds of the Canny algorithm.The experiment results show that the improved Canny algorithm improves the anti-noise ability of the rail surface defect image,improves the edge detection effect of the rail surface defect area,and enhances the robustness of the defect area extraction.(2)A classification algorithm of rail defect image based on the improved ant colony algorithm is proposed.The algorithm analyzes the important steps in the classification of rail defect images,namely the feature extraction of defect images,and it determines a more comprehensive method for feature-selecting functions.In order to solve the problem that the traditional ant colony algorithm is easy to fall into the local optimum,the ant colony algorithm is improved by using the highest similarity of features as the discriminant function.The experimentresults show that the improved ant colony algorithm improves theclassification accuracy when classifying the rail defect images.